Understanding PCA

When to use PCA

Principal Component Analysis (PCA) is an unsupervised multivariate method. It is useful when you want to understand broad structure in the cytokine data without forcing the samples to separate by known group labels.

That makes PCA a good first multivariate step when your question is exploratory:

Do samples cluster naturally?
Are some groups partially separated even before using a supervised model?
Which cytokines appear to drive the strongest sources of variation?

When not to use PCA

PCA is usually not the best first choice when:

your main goal is supervised class separation
the outcome is a numeric response you want to predict
you need formal group-wise statistical testing rather than exploratory structure

What outputs matter most

The PCA workflow in CytokineProfile Shiny can produce several complementary views:

individuals plot
scree plot
loadings plots
biplot
correlation circle

These outputs answer different questions, so it is best to read them together rather than relying on only one figure.

Individuals plot

This is usually the first PCA figure to inspect. Each point is a sample projected into the reduced component space.

Interpret it by asking:

Do samples from the same group tend to cluster together?
Do groups overlap heavily, or is there partial separation?
Are there obvious outliers that may deserve follow-up?

Because PCA is unsupervised, any separation you see here can be especially informative. The method did not use the class labels to force that pattern.

Scree plot

The scree plot shows how much variance is explained by each principal component.

Use it to decide whether:

the first two components already capture a meaningful amount of structure
a third component is worth inspecting
the data are spread across many components rather than dominated by one or two major axes

If the first few components explain only a small fraction of the total variance, then a 2D PCA plot may be visually convenient but biologically incomplete.

Loadings plots and correlation circle

These outputs help answer which cytokines are driving the components.

In general:

larger absolute loadings indicate stronger contribution to a component
cytokines far from the center of a correlation circle contribute more strongly to the displayed components
cytokines pointing in similar directions tend to vary together
cytokines pointing in opposite directions tend to vary in opposing patterns

Which app arguments matter most

The most important PCA settings are:

PCA Comparison Column: decides how the main sample groups are colored or separated visually.
PCA Stratification Column: adds a second visual grouping layer when you want shapes within colors.
Number of Components: determines how many components are computed and emphasized.
Select Colors for PCA Plot (Optional): lets you override the automatic colors.
Draw Ellipse: adds visual confidence regions to the individuals plot.
Plot Style: enables 2D or 3D plotting.
Plotting Symbols: helps distinguish subgroup structure when you use a second grouping column.

For most app users, one of the biggest decisions still happens earlier in Step 2. The preprocessing or transformation applied there can change which cytokines dominate the PCA, even though it is not chosen again on the PCA settings screen.

Common cautions

Keep these limits in mind:

PCA does not test for statistical significance between groups.
Separation in PCA can be influenced by scaling choices and outliers.
A clear individuals plot does not automatically mean the components are easy to interpret biologically.
If group labels are your main outcome of interest, PCA should often be followed by a supervised method rather than treated as the final answer.

How to reproduce the result in the app

Filter the dataset to the samples and cytokines you want to explore.
Choose Principal Component Analysis (PCA).
Set PCA Comparison Column, PCA Stratification Column, and Number of Components.
Decide whether to turn on Draw Ellipse, switch Plot Style, or adjust Plotting Symbols.
Read the individuals plot together with the scree and loadings views.